Current Research Interests:
Research Statement
The proliferation of electronic networks and of client-server technologies has created a new era of software applications for harnessing information. The rapid growth in the number and size of large databases and the application-driven demand to extract knowledge from them has increased the interest in knowledge discovery in databases. Intelligent agents can make information useful by exploring the information environment (e.g. enterprise intranets, financial market data, news-wire services, digital libraries) in order to provide efficient and intelligent access and analysis to essential information, according to the decision context of the user.
Machine learning techniques - reinforcement learning and unsupervised learning in particular - together with distributed problem-solving approaches, can be employed for building intelligent agents for tasks such as process scheduling, data analysis and information routing, filtering and retrieval. These agents are able to operate in real-time in order to satisfy a set of time-dependent goals or motivations; to improve competence in meeting these goals based on experience; and to adapt to unforeseen situations. In addition, the use of the computational market metaphor provides the mechanisms for learning and adaptation amongst such agents as well as a new framework for engineering distributed software systems.
Projects
- SIGMA: Multi-agent systems based on computational markets;
- a multi-agent computational market framework as an integrated methodology for software engineering;
- agents for intelligent workflow design, monitoring and decision support in enterprise information networks;
- the integration of reinforcement learning with unsupervised learning and its application to autonomous agents for planning and control tasks;
- the application of neural networks and of other non-linear modeling and optimization techniques for data analysis and decision-making in finance, manufacturing and medicine;
- the incorporation of pragmatic considerations (resource and accuracy constraints such as time, space, accuracy, cost of testing and cost of errors) as reguralizers in inductive learning systems.